Skip to content
← Tools / Skills
Tooldebpalash/10 trustMetrics-checked

VoiceStudio

Open-source, fully-local voice studio: cloning, dubbing, dictation, and audiobooks.

Visit official site →

About

VoiceStudio (formerly OmniVoice-Studio) is an open-source desktop app by developer Palash Debnath, positioned as a fully-local ElevenLabs alternative. It does zero-shot voice cloning, voice design, cinematic video dubbing, real-time dictation, transcription, and audiobook creation, claiming coverage of 646 languages, all on the user's own hardware with no accounts or per-character billing. It also exposes a local REST/SSE/WebSocket API, an OpenAI-compatible audio API, and an MCP server, with agent skills installable via npx skills add debpalash/VoiceStudio. The app itself is AGPL-3.0, but its default OmniVoice model weights are CC-BY-NC, so commercial use requires a different engine or a separate grant.

What it can do

  • Zero-shot voice cloning from a short reference recording
  • Voice design: describe a voice in plain language and generate it
  • Cinematic video dubbing with timed speech
  • Real-time dictation via a floating widget
  • Transcription, stories, audiobooks, and batch jobs
  • Local models manager: install and switch speech engines and models on disk
  • Local REST/SSE/WebSocket API plus an OpenAI-compatible audio API
  • MCP server for agent-driven voice workflows
  • Headless Docker build for homelab or GPU servers
  • CUDA/ROCm/CPU auto-detect with Apple Silicon Metal acceleration
  • Analytics are opt-in only and can be fully disabled

Setup

  1. Install one of: download the latest Electron release from https://github.com/debpalash/VoiceStudio/releases/latest (macOS .dmg, Windows .exe, Linux .AppImage or .deb), or run curl -fsSL https://voicestudio.sh/install | sh on macOS/Linux (on Windows PowerShell: irm https://voicestudio.sh/install | iex).
  2. Open Voice cloning, choose a voice or add a clean reference recording, enter your text, and generate. The app prompts you to install the required model; check the model license before commercial use.
  3. For agent-driven workflows, run npx skills add debpalash/VoiceStudio and choose the voicestudio skill. Docker users can run the headless web-server build for a homelab or GPU box.

Prerequisites

macOS, Windows, or Linux desktop (Intel Macs: UI only, needs a remote backend); about 5GB free disk for the CPU PyTorch build; GPU optional (CUDA or Apple Silicon Metal acceleration).

Use cases

audiottsvideomedia

Works with

Claude CodeCursorCodexAny agent (.md)

Trust score — 8/10

/10 trustMetrics-checked

70K+ stars and 4,214 commits across 42 releases show strong, sustained adoption for a solo-maintained project, and the codebase is fully public with documented licensing. Points deducted for the default OmniVoice model weights being CC-BY-NC (non-commercial), the active-beta status of the Electron rewrite (Tauri shell retired at 0.5.3), and the real hardware cost of running large models locally.

★ 70,367 GitHub stars

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
9.0/10
Maintenance · recency, releases, issue load
8.0/10
Openness · license + docs
7.0/10
Manual review · sanity read + red-flag check
8.0/10
  • · 70,367 stars, 7,708 forks as of 2026-10-16
  • · AGPL-3.0 app; default OmniVoice model weights are CC-BY-NC — verify model licenses before commercial use
  • · 4,214 commits, 42 releases, solo maintainer; support is Discord plus GitHub issues

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.